Study On Social Network Information Dissemination Prediction Based On User Features

Social network is a complex large-scale relationship network formed by users of the network communicating and interacting with each other.With the tremendous development of mobile internet technology and social media software technology,the factors affecting the information dissemination of social networks have also undergone great changes.The analysis and mining of information dissemination of social networks has important practical significance in hot spot discovery,product marketing,public opinion management and other applications.As an important social network platform,micro-blog has become an important carrier for people to share,acquire and disseminate information with its content introduction,convenient interaction and rapid dissemination.Micro-blog forwarding is an important mechanism for information dissemination on the micro-blog platform,which makes the information explode.Therefore,it is especially necessary to predict the forwarding behavior of users.The challenge of forwarding behavior prediction is how to find more valuable influencing factors to improve predicted performance.This paper deeply analyzes the user features that affect micro-blog forwarding,and summarizes the user extended features set composed of four categories of user influence features,micro-blog topic features,user activity features and user interest features.The analysis results show that micro-blog is highly differentiated by the forwarding rate,the activity of user forwarding,and the strength of interaction between users.However,the features of the number of fans,the number of follow,and the PageRank value of the user are not obvious.Based on the user feature set of this paper,the principle of Naive Bayes algorithm is analyzed,and the construction process of classifier is deduced.In order to make up for the deficiency of the conditional independence hypothesis limitation of naive Bayesian algorithm,this paper uses a new features weighting method to improve the algorithm.At the same time,according to the random characteristics of the training data set,the initial hypothesis of the optimal training subset in the training set is proposed.An incremental optimization method is used to obtain an optimized training data set with higher prediction performance.Conduct a series of experiments on the crawling real social network Sina Micro-Blog.The experimental results show that:(1)Compared with the features set in other studies,using the user extended features set in this paper,the performance index of each classification prediction algorithm has a certain extent,which verifies the effectiveness of the features analysis extraction method.(2)The improved weighted naive Bayesian algorithm has a maximum prediction accuracy of 93%.Compared with the original algorithm,the prediction accuracy is improved by 8%,which achieves the expected effect of the experiment.(3)Through the incremental optimization experiment,a better sample training set is found.Using this set can further improve the performance and effect of the prediction,and verify our initial assumptions.